An MCP server that lets AI assistants control grandMA2 lighting consoles via Telnet. It exposes 41 high-level tools for cue management, fixture control, preset management, executor control, macro editing, appearance assignment, bulk operations, console state queries, show file management, read-back verification, music show workflows, and more through the Model Context Protocol.

• Overview
• Features
• Getting Started
  • Prerequisites
  • Installation
  • Configuration
  • MCP Registration
• Usage
  • MCP Tools
  • Command Builder
  • Direct Telnet Access
• Command Reference
  • Helping Keywords
  • Object Keywords
  • Function Keywords
  • At Keyword
  • Copy and Move
  • Assign
  • Label
  • Appearance
  • Macro Placeholder
• Project Structure
• Development
• Troubleshooting
• License
• Contributing

grandMA2 consoles accept commands over Telnet, but building correct command strings requires knowledge of the MA2 syntax rules -- keyword ordering, preset type mappings, option flags, and object hierarchies. This project handles that complexity in two layers:

1. Command Builder (src/commands/) -- Python functions that construct valid grandMA2 command strings following the official syntax rules. Each function is a thin wrapper that returns a string; it never touches the network.
2. MCP Server (src/server.py) -- A FastMCP server that exposes 41 tools covering cue management, fixture control, presets, executors, global state, labeling, appearance assignment, macro editing, bulk cue operations, sequence playback, console state queries, show file management, read-back verification, music show workflows, and raw commands over stdio transport. It manages a persistent Telnet connection to the console and delegates command construction to the builder layer.
3. Response Parser (src/response_parser.py) -- Pure functions for parsing grandMA2 Telnet output (from List commands) into structured data. Used by read-back tools to return parsed dicts instead of raw text.
4. GMA2Client (src/gma2_client.py) -- A high-level orchestration class that composes multiple command builder calls into workflow-level methods (e.g., build an entire cue list, set up a fixture group with a preset, create song objects for music shows).
5. CommandSequence (src/command_sequence.py) -- A builder-pattern class for composing multiple commands into an ordered batch that can be previewed and executed as a unit.

The command builder covers all grandMA2 command-line keywords organized by category: object keywords (fixtures, channels, groups, presets, cues, sequences, executors), function keywords (store, delete, copy, move, goto, label, assign, and many more), and helping keywords (thru, at, +).

• 41 MCP tools -- Cue management (store/delete/goto, CMD assignment), fixture control (set values, set attributes, clear programmer), preset management (store/apply), executor control (on/off/go/kill/toggle, fader level, sequence assignment), global state (blackout, highlight), object labeling (generic + sequence-scoped cue labeling), appearance assignment (RGB, HSB, hex, source copy), macro line editing, bulk cue operations across sequence ranges (store, label, appearance), sequence playback, console state queries (list groups/cues/presets/variables, read object annotations, generic object query), show file management (save/load/new/list shows), read-back verification (read macro lines, cue info, object labels with structured parsing), music show workflows (create song objects, setup song macros, build set lists), and raw command execution.
• Destructive command safety warnings -- Delete tools include informational warnings about downstream effects (orphaned executor handles, lost cue programming) to help AI assistants understand impact before confirming.
• Complete command builder -- Over 200 Python functions covering all grandMA2 command-line keywords across 30+ modules, each returning a correctly formatted command string.
• High-level client -- GMA2Client provides workflow-level methods: build cue lists, set up fixture groups with presets, quick look programming, batch executor assignments. Uses grandMA2's inline naming syntax to minimize Telnet round-trips.
• Command chaining -- CommandSequence lets you compose multiple commands, preview them, and execute them as a batch.
• Resilient Telnet client -- Built on telnetlib3 with automatic login, persistent connections, connection health checking, auto-reconnection with bounded exponential backoff, command serialization via asyncio.Lock, and graceful shutdown. If the console restarts or the network drops, the client detects the failure and reconnects transparently before the next command.
• Connection error surfacing -- MCP tools catch connection failures and return human-readable error messages instead of unhandled exceptions, so the AI assistant knows when commands are not reaching the console.
• Configurable transport -- Supports stdio (default, single client) and streamable-http (multi-client, web-based access) transports. HTTP host and port are configurable via environment variables.
• Configurable via environment -- Host, port, user, password, and transport settings set through .env or environment variables.

• Python >= 3.12
• A grandMA2 console (or onPC) with Telnet access enabled
• uv (recommended) or pip

On macOS, if you need a Telnet client for manual testing:

brew install telnet

git clone <repository-url>
cd gma2-mcp

Using uv:

uv sync

Using pip:

python -m venv .venv
source .venv/bin/activate
pip install -e .

Copy the template and fill in your console details:

cp .env.template .env

GMA_HOST should be set to the IP address of the machine running grandMA2 onPC. You can find this in the onPC network settings or by checking the machine's network configuration.

Port 30000 is the standard command port. Port 30001 is read-only (log output).

To enable HTTP transport for web-based or multi-client access:

MCP_TRANSPORT=streamable-http
MCP_HOST=0.0.0.0    # bind to all interfaces (default: 127.0.0.1)
MCP_PORT=3000        # custom port (default: 8000)

When using streamable-http, concurrent commands from multiple clients are automatically serialized via an internal lock to prevent interleaved Telnet commands, since the grandMA2 console processes commands sequentially.

Add the server to your MCP client configuration.

Claude Desktop (~/Library/Application Support/Claude/claude_desktop_config.json on macOS):

{
  "mcpServers": {
    "gma2": {
      "command": "uv",
      "args": [
        "--directory",
        "/path/to/gma2-mcp",
        "run",
        "python",
        "-m",
        "src.server"
      ],
      "env": {
        "GMA_HOST": "2.0.0.1",
        "GMA_USER": "administrator",
        "GMA_PASSWORD": "admin"
      }
    }
  }
}

Without uv -- point directly to the virtualenv Python:

{
  "mcpServers": {
    "gma2": {
      "command": "/path/to/gma2-mcp/.venv/bin/python",
      "args": ["-m", "src.server"],
      "cwd": "/path/to/gma2-mcp",
      "env": {
        "GMA_HOST": "2.0.0.1",
        "GMA_USER": "administrator",
        "GMA_PASSWORD": "admin"
      }
    }
  }
}

Claude Code (CLI / IDE extensions)

Claude Code manages MCP servers through its own configuration, separate from Claude Desktop. You can set it up using either the CLI command or by editing the settings file directly.

Option 1: Using the claude mcp add command (recommended)

Run the following command in your terminal:

claude mcp add gma2 \
  -e GMA_HOST=2.0.0.1 \
  -e GMA_USER=administrator \
  -e GMA_PASSWORD=admin \
  -- uv --directory /path/to/gma2-mcp run python -m src.server

This registers the MCP server with Claude Code. The -e flags set environment variables that the server reads at startup. Replace /path/to/gma2-mcp with the actual path to your cloned repository, and adjust the GMA_HOST, GMA_USER, and GMA_PASSWORD values to match your console.

To register the server for a specific project only (instead of globally), add the -s project flag:

claude mcp add gma2 -s project \
  -e GMA_HOST=2.0.0.1 \
  -e GMA_USER=administrator \
  -e GMA_PASSWORD=admin \
  -- uv --directory /path/to/gma2-mcp run python -m src.server

Option 2: Editing the settings file manually

Add the server entry to your Claude Code settings file. The file location depends on the scope:

• User-level (available in all projects): ~/.claude/settings.json
• Project-level (available only in the current project): .claude/settings.json in the project root

{
  "mcpServers": {
    "gma2": {
      "command": "uv",
      "args": [
        "--directory",
        "/path/to/gma2-mcp",
        "run",
        "python",
        "-m",
        "src.server"
      ],
      "env": {
        "GMA_HOST": "2.0.0.1",
        "GMA_USER": "administrator",
        "GMA_PASSWORD": "admin"
      }
    }
  }
}

Verifying the setup

After registering the server, verify it is working:

# List all registered MCP servers
claude mcp list

# Check the server details
claude mcp get gma2

When you start a new Claude Code session, the server starts automatically. You can confirm the tools are available by asking Claude to list its MCP tools or by directly requesting a grandMA2 operation (e.g., "toggle blackout on the console").

Managing the server

# Remove the server
claude mcp remove gma2

# Re-add with different settings
claude mcp add gma2 \
  -e GMA_HOST=192.168.1.100 \
  -- uv --directory /path/to/gma2-mcp run python -m src.server

The server exposes 41 tools:

The query tools use send_command_with_response() to capture the grandMA2 Telnet output and return raw text. The grandMA2 manual does not document the exact Telnet wire format for List and Info responses, so these tools return unprocessed console output that the AI can interpret directly.

• list_groups, list_cues, list_presets -- Use the grandMA2 List keyword to retrieve show data. Accept optional ID ranges (e.g., list_groups(group_id=1, end_group_id=10)). list_presets validates the preset type against the 9 known types before sending.
• get_cue_annotation, get_group_annotation -- Use the grandMA2 Info keyword, which reads user-added descriptive text annotations on objects (set via Info Group 3 "some note" on the console). These do NOT return object properties like fade times or fixture composition -- use list_cues or list_groups for that.
• list_variables -- Routes to ListVar (show variables) or ListUserVar (user variables) based on the variable_type parameter. Supports an optional filter pattern.
• query_object -- Generic query for any grandMA2 object type not covered by the dedicated tools. Supports mode="list" (default) and mode="annotation".

All query tools return a descriptive message instead of an empty string when the console returns no data.

• save_show_tool -- Saves the current show file. Accepts an optional show_name to save as a specific name. Uses /noconfirm to suppress the overwrite confirmation popup that would block Telnet.
• load_show_tool -- Loads a show file by name. This is a destructive operation: unsaved changes to the current show are lost. Accepts save_first=True to save the current show before loading. Uses /noconfirm to suppress the GUI popup.
• new_show_tool -- Creates a new empty show. Same destructive behavior and save_first option as load_show_tool.
• list_shows_tool -- Lists available show files on the console's selected drive. Accepts an optional filter pattern.

The read-back tools use send_command_with_response() to capture grandMA2 List command output and parse it into structured dicts via src/response_parser.py. Each tool returns a dict with parsed fields plus a raw_response field for debugging. If the response format is unrecognized, the tools return parsed: False with the raw output.

• read_macro_lines -- Sends List Macro {pool}.{id} and parses each line's number and CMD content. Returns {"macro_id": N, "parsed": True, "lines": [{"line_number": 1, "cmd": "..."}], "raw_response": "..."}.
• read_cue_info -- Sends List Cue {id} Sequence {seq} and extracts label, CMD, and fade time. Returns {"sequence_id": N, "cue_id": N, "label": "...", "cmd": "...", "fade": "...", "raw_response": "..."}.
• read_object_label -- Sends List {type} {id} and extracts the object's name/label. Works with any pool object type (sequence, page, group, etc.). For macros, the object_id must be pool-qualified (e.g., "1.5" for Macro 5 in Pool 1).

High-level tools that automate common multi-step patterns for music show programming. These delegate to GMA2Client workflow methods.

• create_song_objects -- Creates and labels a Sequence + Page pair with the same ID and name. This is the standard pattern for per-song programming (e.g., create_song_objects(song_id=101, song_name="Opening+Childhood") sends 2 commands).
• setup_song_macro -- Creates a labeled macro with a SetVar command on line 1 for tracking the current song. Accepts an optional var_name (default: $song). Sends 3 commands: store macro, label, assign CMD.
• build_set_list -- Creates a set-list sequence with cues linked to song macros. For each song, stores a cue with the song name as label and assigns Macro {id} as the cue CMD. Accepts a list of {cue_id, macro_id, name} dicts.

delete_show is intentionally not exposed as an MCP tool -- it is irreversible and too dangerous for AI-initiated operations. The Backup command is also excluded because it opens a GUI menu on the console (per grandMA2 Manual p368), which is not functional over Telnet. Use save_show_tool with a specific name as the practical backup mechanism.

The src/commands/ module can be used independently to build command strings:

from src.commands import fixture, at_full, store_group, label_group

fixture(1, end=10)              # "fixture 1 thru 10"
at_full()                       # "at full"
store_group(5)                  # "store group 5"
store_group(5, name="Front Wash")  # 'store group 5 "Front Wash"' (inline naming)
label_group(5, "Front Wash")    # 'label group 5 "Front Wash"' (separate label command)

The GMA2Client class provides high-level workflow methods that compose multiple commands. These methods use grandMA2's inline naming syntax to minimize the number of Telnet round-trips:

from src.gma2_client import GMA2Client

async with GMA2Client.create("192.168.1.100") as client:
    # Build a cue list with names and fade times
    # Uses inline naming: 1 command per named cue (store cue N "Name"),
    # plus 1 command per fade time
    await client.build_cue_list(1, [
        {"id": 1, "name": "Preset", "fade": 0},
        {"id": 2, "name": "Look 1", "fade": 3.0},
        {"id": 3, "name": "Blackout", "fade": 2.0},
    ])

    # Select fixtures, store as named group, apply a preset (3 commands)
    # Uses inline naming: store group N "Name" in a single command
    await client.setup_group_with_preset(
        fixtures=(1, 10), group_id=1,
        group_name="Front Wash", preset_type="color", preset_id=3,
    )

    # Quick look: set fixtures to a value, optionally store as cue
    await client.quick_look(fixtures=(1, 20), value=75, store_as_cue=5)

    # Batch assign sequences to executors
    await client.assign_sequences_to_executors([(1, 1), (2, 2), (3, 3)])

    # Clone fixture programming (with /noconfirm for telnet)
    await client.clone_fixtures(source_fixture=1, target_fixture=11,
                                source_end=5, target_end=15, mode="overwrite")

    # Apply effect to a group with parameters
    await client.setup_effect_on_group(group_id=1, effect_id=5,
                                       bpm=120, form="sin", high=100, low=0)

    # Set up a full executor page
    await client.setup_executor_page(page=1, assignments=[
        {"executor_id": 1, "sequence_id": 1, "label": "Wash", "fader_level": 80},
        {"executor_id": 2, "sequence_id": 2, "label": "Spots"},
    ])

    # Label multiple objects at once
    await client.batch_label("group", {1: "Wash", 2: "Spots", 3: "Beams"})

    # Create and optionally run a macro
    await client.create_and_run_macro(
        macro_id=10, commands=["Go Sequence 1", "Go Sequence 2"],
        name="Start Show", run=True,
    )

    # Music show workflows: create song objects (sequence + page pair)
    await client.create_song_objects(song_id=101, song_name="Opening+Childhood")

    # Set up a song macro with SetVar on line 1
    await client.setup_song_macro(macro_id=101, song_name="Opening+Childhood")

    # Build a full set list with cue-to-macro links
    await client.build_set_list(
        sequence_id=100, sequence_name="Set List",
        songs=[
            {"cue_id": 1, "macro_id": 101, "name": "Opening+Childhood"},
            {"cue_id": 2, "macro_id": 102, "name": "Nostalgia"},
            {"cue_id": 3, "macro_id": 103, "name": "Finale"},
        ],
    )

The CommandSequence class lets you compose multiple commands and execute them as a batch:

from src.command_sequence import CommandSequence
from src.commands import fixture_at, store_group, at_full

seq = CommandSequence()
seq.add(fixture_at(1, 50)).add(at_full()).add(store_group(1))

# Preview before sending
print(seq.preview())  # ['fixture 1 at 50', 'at full', 'store group 1']

# Execute all commands
result = await seq.execute(client)
# {'commands_sent': [...], 'count': 3, 'success': True}

For manual testing via the Makefile:

make server    # Connect to grandMA2 command port (30000)
make log       # Connect to log output port (30001)

To exit a Telnet session: press Ctrl + ], then type quit.

uv run python -m src.server

The command builder follows the official grandMA2 command-line syntax. The basic pattern is [Function] [Object]. All objects have a default function, and most functions have a default object type. Objects are arranged in a hierarchical tree.

Used to create relations between functions and objects.

Allocate objects in the show file. Usually combined with numbers, IDs, names, or labels.

Perform a task or function, often followed by the object they apply to.

At functions as both a Function Keyword and a Helping Keyword.

Copy creates duplicates. Move relocates objects (swaps if the destination is occupied).

Copy options: overwrite, merge, status, cueonly, noconfirm

Defines relationships between objects, patching, and property assignment.

Assign options: break_, multipatch, reset, x, y, noconfirm, special, cue_mode, password

Gives names to objects. Numbers in names auto-enumerate for ranges.

Changes frame colors of pool objects and background colors of cues.

The @ character (distinct from the At keyword) is used as a placeholder for user input in macros.

gma2-mcp/
├── src/
│   ├── commands/               # Command builder module
│   │   ├── __init__.py         # Public API exports
│   │   ├── constants.py        # PRESET_TYPES, store option sets
│   │   ├── helpers.py          # Internal helper functions
│   │   └── functions/          # Function keywords by category
│   │       ├── advanced_edit.py    # Flip, CircularCopy, Import, Export
│   │       ├── assignment.py       # Assign keyword functions
│   │       ├── blackout.py         # Blackout, Freeze, Highlight, Solo
│   │       ├── blind.py            # Blind, Preview modes
│   │       ├── call.py             # Call keyword
│   │       ├── conditionals.py     # EndIf, IfActive, IfOutput, Or, With
│   │       ├── crossfade.py        # Crossfade, ManualXFade
│   │       ├── cue_timing.py       # Delay, Fade timing
│   │       ├── edit.py             # Copy, Move, Delete, Remove
│   │       ├── effect.py           # Effect, EffectBPM, EffectForm
│   │       ├── executor_control.py # Off, On, Kill, Flash, Swop
│   │       ├── fixture_control.py  # Align, Fix, Locate, Next, Prev
│   │       ├── flash_swop_ext.py   # FlashGo, SwopGo, StoreLook
│   │       ├── helping.py          # Helping keywords (Thru, +)
│   │       ├── info.py             # List and Info queries
│   │       ├── intensity.py        # Full, Zero, Load, Learn
│   │       ├── labeling.py         # Label and Appearance
│   │       ├── list_ext.py         # ListShows, ListOops, ListVar
│   │       ├── macro.py            # Macro placeholder functions
│   │       ├── matricks.py         # MAtricks keywords
│   │       ├── midi.py             # MIDI control functions
│   │       ├── navigation.py       # Down, Up, NextRow, Search
│   │       ├── network.py          # Remote, Telnet, JoinSession
│   │       ├── park.py             # Park/Unpark functions
│   │       ├── playback.py         # Go, GoBack, Goto, DefGo
│   │       ├── programmer.py       # Block, Clone, Default, Update
│   │       ├── rate_speed.py       # Rate, Speed control
│   │       ├── rdm.py              # RDM functions
│   │       ├── selection.py        # SelFix and Clear functions
│   │       ├── show_data.py        # CrashLog, Lua, PSR, Thru
│   │       ├── step_timing.py      # SnapPercent, StepFade, FadePath
│   │       ├── store.py            # Store functions
│   │       ├── system.py           # Backup, Setup, Shutdown, Login
│   │       ├── values.py           # At and value setting
│   │       └── variables.py        # Variable functions
│   ├── command_sequence.py     # CommandSequence for multi-command batching
│   ├── gma2_client.py          # High-level workflow orchestration client (15 methods)
│   ├── response_parser.py      # Parse grandMA2 Telnet output into structured data
│   ├── server.py               # MCP server (FastMCP, 41 tools, configurable stdio/streamable-http transport)
│   └── telnet_client.py        # Async Telnet client with health check, auto-reconnect, command lock, state tracking
├── tests/                      # Pytest test suite (one file per module)
├── doc/                        # grandMA2 user manual (PDF)
├── connect.sh                  # Telnet connection script with auto-login
├── Makefile                    # Utility commands (server, log, test)
├── pyproject.toml              # Project metadata and dependencies
└── .env.template               # Environment variable template

make test

Or directly:

uv run pytest -v

Run a specific test file:

uv run pytest tests/test_playback.py -v

The project has five layers:

1. Telnet Client (src/telnet_client.py) -- Low-level async Telnet communication with the console. Includes ConnectionState tracking (DISCONNECTED/CONNECTING/CONNECTED/RECONNECTING), health check probing, auto-reconnection with bounded exponential backoff (configurable retries and delay), health check TTL to skip redundant probes during rapid command sequences, asyncio.Lock-based command serialization for concurrent access safety, and graceful shutdown via the server lifespan hook.
2. Command Builder (src/commands/) -- Pure functions that construct command strings. No network access.
3. Response Parser (src/response_parser.py) -- Pure functions that parse grandMA2 Telnet List command output into structured dicts. Returns parsed: False on unrecognized formats.
4. Orchestration (src/gma2_client.py, src/command_sequence.py) -- High-level workflow methods (15 methods including music show workflows) and command batching that compose builders + telnet.
5. MCP Server (src/server.py) -- FastMCP server that exposes 41 tools over configurable transport (stdio or streamable-http), connecting the builder to the Telnet client.

All console communication goes through the Telnet client layer.

Connection fails -- Verify the console IP and port. Ensure Telnet is enabled on the console. Test manually with make server.

Connection drops mid-session -- The Telnet client auto-reconnects with up to 3 retries (exponential backoff: 1s, 2s, 4s). If all retries fail, tools return a clear error message. Check that the console is running and reachable.

Authentication errors -- Check username/password in .env. Confirm the user account exists on the console. Login is case-sensitive.

Fixture setup locked by stale connection -- If a previous session crashed without calling disconnect(), the console may still hold a Telnet lock. Restart the console or wait for its session timeout. The MCP server now calls disconnect() on shutdown to prevent this.

Command not working -- Verify the syntax against the grandMA2 manual. Ensure referenced objects (fixtures, groups, presets) exist in the current show file.

This project is licensed under the Apache License 2.0. See LICENSE for details.

1. Fork the repository
2. Create a feature branch (git checkout -b feature/name)
3. Commit your changes
4. Push to the branch and open a Pull Request

Run tests before submitting: make test